OBJECTIVES: EQ-5D data are often summarised by the EQ-5D index. The distribution of the EQ-5D-3L index shows two distinct groups (Parkin et al., 2014). This might reflect the actual distribution of ill health, but might also be an artefact of how the EQ-5D-3L Index is constructed. There has been little work to explore the distribution of EQ-5D-5L index. In this research project we explore whether or not the EQ-5D-5L index distribution also demonstrates clustering. We test the extent to which clustering of EQ-5D-5L profile data is a driver of an optimal number of index clusters. METHODS: Data are available from Cambridgeshire Community Services NHS's electronic patient records data warehouse, with 30284 patient observations across three patient groups: community rehabilitation services (N=6919); musculoskeletal therapy services (N=19999); and nursing services (N=3366). It includes patients' EQ-5D-5L profiles and EQ-5D index before treatment. Kmeans cluster method and the Calinski–Harabasz pseudo-F index stopping rule are used to search for the optimal number of clusters. The robustness of the solution to the choice of initial value is tested. We examine the distribution of 1730 EQ-5D-5L profiles (out of the total available 3125 profiles) to check whether these explain clustering of the 5L index. RESULTS: Clustering within the 5L index distribution is suggested by both methods and is observed across all the three patient groups. The solution to optimal number of clusters is not robust beyond 4 clusters. The profile data is not a clear driver for the 5L index clusters, in particular for profiles with misery scores in the middle. CONCLUSIONS: Multi-modality characteristics of the 5L index distribution are found in patients' data. The profile data alone cannot distinguish between different index clusters. Looking ahead, future research will investigate the potential for the profile clusters to aid in service design and outcome measurement.